With the increasing demand for high-bandwidth in private cloud, public cloud data centers and communication service providers, 25G and 100G experienced significant uptake. Until now, most server vendors have already provided servers with 25GbE NICs as the standard I/O (input/output) options. The Ethernet speed transitions has been increased from 10G to 25G, 100G or even higher. Although 1G, 10G and 40G still represent a significant share of the enterprise market’s Ethernet ports, the future demand for 25G and 100G will be stronger than ever, because high bandwidth is undeniably pushing the data center towards higher scalability and flexibility.
Why will the data center ushers in 25G?
The data center is expanding at an unprecedented speed, driving the need for higher bandwidth between the server and the switch. In order to cater to this trend, the access network has been gradually upgraded from 10G to 25G. This also provides a high-density, low-cost and low-power solution for the connection between servers and ToR switches.
Evolutionary Path of 25G
Since the 25GbE was initially proposed in 2014, companies such as Google, Microsoft, Arista Networks, Broadcom, and Mellanox, etc. have been pushing the development of the 25G Ethernet standard, with the intention of achieving 25G top-of-rack server networking. Up to till now, 25G is rapidly spreading in the market. In the future, 25G will provide a comprehensive solution for the connection between servers and switches. The following figure 1 shows the key milestones of 25G Ethernet from 2014 to 2018.
Figure-1: Evolutionary Path of 25GbE
The advantages of 25G
Before the 25G Ethernet specification was released, the network upgrade methods of data centers like enterprises and operators generally use 10G-40G upgrade path. With the emergence of 25G Ethernet, 25G-100G upgrade path has gained more applications with the advantages of low cost, low power consumption and high density, which has promoted the rapid development of 100G Ethernet. Let’s take a look at the differences between 10G, 25G, 40G and which upgrade method is superior.
Figure-2: 10G-40G vs 25G-100G Upgrade Path
Compared with 10G, 25G provides higher performance bandwidth
In the current data center, the network connection between servers and switches are generally selected between 10G and 25G. Compared with 10G, 25G has made progress based on 10G, providing higher bandwidth and performance. The emergence of 25G can make the data center based on the existing network architecture, without adding any cable interconnection, to support higher-speed (over 10G) transmission, to meet the future network’s demand for higher bandwidth and upgrade the network more convenient and easy. The wiring infrastructure required for 25G and 10G transmission is basically the same, which can effectively avoid expensive costs and the complexity of rewiring, making network upgrades more convenient.
In addition, 25G uses single-lane 25G SerDes technology similar in operation to 10G SFP, with backward compatibility, significantly reducing power consumption and costs, helping data center operators save CapEx (capital expenditures) and OpEx (operational expenditures).
Insert 25G SFP28 transceiver module into 10G SFP+ port, what speed will we get?
In theory, the 25G SFP28 is backward compatible with 10G SFP+ port, and its rate can reach 10Gb/s. However, this method of use is not necessarily suitable for all brand switches and transceivers. Considering the limitations of NICs and switch ports, it is generally not recommended.
Compared with 40G, 25G is more in line with high density requirements
25G makes the Ethernet specification between NICs and swatches interoperable, providing greater port density and lower cost per unit of bandwidth connection of rack servers. Compared with 40G, it can provide higher port and system density, and maximize the I/O performance and exchange capacity of the switch. Judging from the number of connected switches, since 25G uses single-channel and 40G uses four transmission channels, 25G can connect more switches. In addition, compared with the existing 10G and 40G, the transmission performance of 25G and 50G is improved by 2.5 times.
For big enterprises, the port density of servers largely determines the cost of cabling and switch infrastructure in the entire system. Therefore, the cost of upgrading from 25G to 100G is relatively low compared to 40G. Because the 25G-100G upgrade path, the switch port is fully utilized, effectively reducing the cost of bandwidth.
Having upgraded to a 40G network, is it still necessary to deploy a 25G network?
Compared with 40G equipment, 25G equipment is more expensive, so is it necessary to deploy 25G systems if it has been upgraded to 40G? Due to the rationalization of the cost of the 25G channel, the future 25G is definitely an important path for upgrading 10G to 100G or higher. If you need to increase the baud rate (signal transmission rate) or plan to upgrade the network to a higher speed (100G, 200G, 400G), you must deploy 25G systems. If there is no demand, then you do not need to deploy 25G systems.
Looking at the prospect of 25G from the future 100G, 200G, 400G data center
At present, 25G servers and 100G switches have become ubiquitous in hyperscale data centers. They have gradually replaced the previous 10G servers and 40G switches. This speed migration has increased the throughput of the entire system by 2.5 times, while reducing incremental costs. As the Ethernet industry continues to innovate and lay a path to higher networking speeds, the 25G-100G upgrade has become an important path for data centers.
25G provides more possibilities for 50G
As we all know, 25G compared to 10G only needs one channel to provide 2.5 times bandwidth. In the future, 50G can also provide 1.25 times bandwidth through one channel compared to 40G. Currently 50G has been proposed as the basis of 100G, 200G, 400G upgrade, but the implementation of the 50G Ethernet standard will still take some time.
25G will provide more possibilities for 50G. Since the implementation of 50G Ethernet can be based on two 25G channels, it will become an alternative to the current use of four 10G channels to reach 40G, reducing the cost of network equipment in the data center by reducing channel. . In the future, the network upgrade path may evolve from the traditional 10G-40G-100G to 10G-25G-50G-100G. In any case, data center upgrades to 50G or 100G networks through multiple 25G channels will become simpler and more economical.
Figure-3: Current vs Future High Speed Ethernet Upgrade Mode
25G lay a path to 200G, 400G upgrade
25G, 50G, 100G architecture has higher flexibility, so it is usually used as a solution for large data centers, paving the way for later 200G, 400G upgrades. At present, high-end enterprises and large data centers are changing towards this, effectively promoting the implementation of large data centers and interconnection between data centers. Now more and more suppliers on the market are committed to the research and development of 200G, 400G optical devices, and some of these products have been successfully put into use. The realization of 100G Ethernet is based on the development of 25G, 50G. Similarly, the future 200G, 400G upgrade will be based on 100G. The following table-1 lists the upgrade paths from 25G, 50G, 100G to 200G, 400G.
The future demand for higher speeds and performance in data centers will never stop. Looking back on the evolution of 25G over the past few years, you will find that the emergence of 25G is a milestone in the expansion of the bandwidth and channel capacity of next-generation data center networks. The 25G-100G upgrade provides higher bandwidth and port density, reduces power consumption and cost, subverts the traditional 10G-40G network, improves the efficiency of the data center, and lays a path to 200G, 400G upgrade. Let us wait and see how the continuous development and innovation of Ethernet will promote a new round of data center transformation!